Kernel based algorithms such as support vector machines have achieved considerable success in various problems in the batch setting where all of the training data is available in advance. Support vector machines combine the so-called kernel trick with the large margin idea. There has been little use of these methods in an online setting suitable for real-time applications. In this paper we consider online learning in a Reproducing Kernel Hilbert Space. By considering classical stochastic gradient descent within a feature space, and the use of some straightforward tricks, we develop simple and computationally efficient algorithms for a wide range of problems such as classification, regression, and novelty detection. In addition to allowing the exploitation of the kernel trick in an online setting, we examine the value of large margins for classification in the online setting with a drifting target. We derive worst case loss bounds and moreover we show the convergence of the hypothesis to the minimiser of the regularised risk functional. We present some experimental results that support the theory as well as illustrating the power of the new algorithms for online novelty detection. In addition

Markov networks are widely used in a wide variety of applications, in problems ranging from computer vision, to natural language, to computational biology. In most current applications, even those that rely heavily on learned models, the structure of the Markov network is constructed by hand, due to the lack of effective algorithms for learning Markov network structure from data. In this paper, we provide a computationally effective method for learning Markov network structure from data. Our method is based on the use of L1 regularization on the weights of the log-linear model, which has the effect of biasing the model towards solutions where many of the parameters are zero. This formulation converts the Markov network learning problem into a convex optimization problem in a continuous space, which can be solved using efficient gradient methods. A key issue in this setting is the (unavoidable) use of approximate inference, which can lead to errors in the gradient computation when the network structure is dense. Thus, we explore the use of different feature introduction schemes and compare their performance. We provide results for our method on synthetic data, and on two real world data sets: modeling the joint distribution of pixel values in the MNIST data, and modeling the joint distribution of genetic sequence variations in the human HapMap data. We show that our L1-based method achieves considerably higher generalization performance than the more standard L2-based method (a Gaussian parameter prior) or pure maximum-likelihood learning. We also show that we can learn MRF network structure at a computational cost that is not much greater than learning parameters alone, demonstrating the existence of a feasible method for this important problem. 1

An accurate mapping of traffic to applications is important for a broad range of network management and measurement tasks. Internet applications have traditionally been identified using well-known default server network-port numbers in the TCP or UDP headers. However this approach has become increasingly inaccurate. An alternate, more accurate technique is to use specific application-level features in the protocol exchange to guide the identification. Unfortunately deriving the signatures manually is very time consuming and difficult. In this paper, we explore automatically extracting application signatures from IP traffic payload content. In particular we apply three statistical machine learning algorithms to automatically identify signatures for a range of applications. The results indicate that this approach is highly accurate and scales to allow online application identification on high speed links. We also discovered that content signatures still work in the presence of encryption. In these cases we were able to derive content signature for unencrypted handshakes negotiating the encryption parameters of a particular connection.

Data integration systems offer a uniform interface to a set of data sources. Despite recent progress, setting up and maintaining a data integration application still requires significant upfront effort of creating a mediated schema and semantic mappings from the data sources to the mediated schema. Many application contexts involving multiple data sources (e.g., the web, personal information management, enterprise intranets) do not require full integration in order to provide useful services, motivating a pay-as-you-go approach to integration. With that approach, a system starts with very few (or inaccurate) semantic mappings and these mappings are improved over time as deemed necessary. This paper describes the first completely self-configuring data integration system. The goal of our work is to investigate how advanced of a starting point we can provide a pay-as-you-go system. Our system is based on the new concept of a probabilistic mediated schema that is automatically created from the data sources. We automatically create probabilistic schema mappings between the sources and the mediated schema. We describe experiments in multiple domains, including 50-800 data sources, and show that our system is able to produce high-quality answers with no human intervention.

Several algorithms have been proposed to learn to rank entities modeled as feature vectors, based on relevance feedback. However, these algorithms do not model network connections or relations between entities. Meanwhile, Pagerank and variants find the stationary distribution of a reasonable but arbitrary Markov walk over a network, but do not learn from relevance feedback. We present a framework for ranking networked entities based on Markov walks with parameterized conductance values associated with the network edges. We propose two flavors of conductance learning problems in our framework. In the first setting, relevance feedback comparing node-pairs hints that the user has one or more hidden preferred communities with large edge conductance, and the algorithm must discover these communities. We present a constrained maximum entropy network flow formulation whose dual can be solved efficiently using a cutting-plane approach and a quasi-Newton optimizer. In the second setting, edges have types, and relevance feedback hints that each edge type has a potentially different conductance, but this is fixed across the whole network. Our algorithm learns the conductances using an approximate Newton method.

Abstract. In this paper we unify divergence minimization and statistical inference by means of convex duality. In the process of doing so, we prove that the dual of approximate maximum entropy estimation is maximum a posteriori estimation. Moreover, our treatment leads to stability and convergence bounds for many statistical learning problems. Finally, we show how an algorithm by Zhang can be used to solve this class of optimization problems efficiently. 1

Abstract. We present a unified and complete account of maximum entropy distribution estimation subject to constraints represented by convex potential functions or, alternatively, by convex regularization. We provide fully general performance guarantees and an algorithm with a complete convergence proof. As special cases, we can easily derive performance guarantees for many known regularization types, including ℓ1, ℓ2, ℓ 2 2 and ℓ1 + ℓ 2 2 style regularization. Furthermore, our general approach enables us to use information about the structure of the feature space or about sample selection bias to derive entirely new regularization functions with superior guarantees. We propose an algorithm solving a large and general subclass of generalized maxent problems, including all discussed in the paper, and prove its convergence. Our approach generalizes techniques based on information geometry and Bregman divergences as well as those based more directly on compactness. 1

We study the problem of modeling species geographic distributions, a critical problem in conservation biology. We propose the use of maximum-entropy techniques for this problem, specifically, sequential-update algorithms that can handle a very large number of features. We describe experiments comparing maxent with a standard distribution-modeling tool, called GARP, on a dataset containing observation data for North American breeding birds. We also study how well maxent performs as a function of the number of training examples and training time, analyze the use of regularization to avoid overfitting when the number of examples is small, and explore the interpretability of models constructed using maxent. 1.

We study the problem of maximum entropy density estimation in the presence of known sample selection bias. We propose three bias correction approaches. The first one takes advantage of unbiased sufficient statistics which can be obtained from biased samples. The second one estimates the biased distribution and then factors the bias out. The third one approximates the second by only using samples from the sampling distribution. We provide guarantees for the first two approaches and evaluate the performance of all three approaches in synthetic experiments and on real data from species habitat modeling, where maxent has been successfully applied and where sample selection bias is a significant problem. 1

The problem of maximum-likelihood learning of the Markov tree structure of an unknown distribution from samples is considered when the distribution is Markov on a tree. Large-deviation analysis of the error in estimation of the set of edges of the tree is considered. Necessary and sufficient conditions are provided to ensure that this error probability decays exponentially. These conditions are based on the mutual information between each pair of variables being distinct from that of other pairs. The rate of error decay, which is the error exponent, is derived using the large-deviation principle. For a discrete distribution, the error exponent is approximated using Euclidean information theory, and is given by a ratio, interpreted as the signal-to-noise ratio (SNR) for learning. Extensions to the Gaussian case are also considered.